Key Takeaways

• Power users of AI coaching are 28% more likely to move up a performance rating: ADI's analysis found that employees who used Nadia AI coaching with high frequency, broad topic coverage, and deep conversation quality were significantly more likely to advance to a higher performance tier compared to casual users or non-users.
• Manager adoption is a force multiplier: When a manager is a power user of AI coaching, their direct reports are two times more likely to become active users themselves. At ADI, 77% of managers engaged with Nadia versus 61% of individual contributors — and teams whose managers used Nadia had a 67% engagement rate compared to 33% for those whose managers did not.
• AI coaching can close rather than widen equity gaps: ADI found that women — who studies show are less likely to use generative AI when controlling for all factors — represented nearly gender parity (47%) among Nadia's top power users, meaning women at ADI were 68% more likely to be in the high-frequency power user group relative to their share of the workforce.
• Adoption metrics alone are insufficient — conversation quality matters: The Power User Index incorporates volume, breadth of use cases, and conversation depth to distinguish meaningful AI engagement from superficial experimentation. An employee who opens a tool once a quarter looks nothing like one who uses it to think through high-stakes decisions regularly.
• Workforce bifurcation is a real and measurable risk: When power users gain even a modest productivity amplification advantage over casual users, the absolute performance gap compounds over time. HR leaders who do not actively engineer equitable AI adoption across their organizations risk accelerating internal performance divides.
• Invitations drive engagement more effectively than mandates: Jennifer Carpenter's key lesson: invite employees to explore AI coaching rather than requiring it. People learn when they feel engaged and empowered, not when they are on their back foot. ADI's approach of starting with use-case relevance — reducing the pain of tasks like goal setting and self-reviews — drove organic adoption.

Full Session Transcript

Introducing the Research: 45,000 AI Coaching Sessions at ADI

Das Rush: As they make their way, I want to highlight — because people will claim research in a lot of different ways — the depth of what these two have done. I started familiarizing myself with it in the last couple of months, and the more I started asking questions about what they were looking at, the more I was like, oh my goodness, you have really dug in here. Across 45,000 coaching sessions in over 14 months, working together with our data team at Valence, they've come up with some emerging findings they're going to share today. To get us started, Jennifer, at Valence, we love context. I wanted to lay out the context before we dug into the results for what AI coaching has been at ADI and how you've been using it. Tell the story of your journey, first with Nadia, and then we'll talk about what you're seeing in the data.

Jennifer Carpenter: At ADI, we introduced Nadia in October 2024. Nadia was released in 2023, so we were among the early adopters. And if you can remember back just a short time ago, even in that timeframe, it was new to be talking about agentic AI. I had to explain that Nadia is like an agent — it is not like ChatGPT. We were thinking about ways to get people into the gym to do reps with AI, because there was a lot of lack of understanding. Some people were experimenting, some were not. We thought AI coaching was a really low-risk use case to get people to try it, find value, and see the art of the possible about what it would be like working with an agent or thought partner. That was a big point.

We are a global company. One thing that really appealed to me as a long-time talent leader: it's the first time in my career that I can put my hand on heart and say we're building a truly inclusive product in that it speaks so many languages. We had employees very early on saying, 'I can speak to Nadia in my mother tongue,' whether that be Hindi or — we have a large population in Ireland where a woman said, 'She speaks Gaelic.' For those evaluating any AI coaching product, the ability to unlock the possibility of helping a workforce through the inclusivity of engaging in whatever language you're comfortable with was really important to us.

What AI Coaching for Enterprises Looks Like in Practice

AI coaching for enterprises involves deploying an AI agent — like Valence's Nadia — that employees can use to think through high-stakes decisions, prepare for difficult conversations, set goals, and develop leadership skills. At Analog Devices, Inc. (ADI), 45,000 AI coaching sessions were conducted over 14 months across a global workforce. Employees could engage in their native language, including Hindi and Irish Gaelic, making the tool meaningfully inclusive from day one.

Defining the Power User Index for AI Coaching Adoption

Das Rush: Prasad, you spent your career studying teams and performance. Why did this project speak to you? Why is this an important data project for understanding how AI usage is impacting the workforce?

Prasad Setty: It's more than a project — it's an initiative. We need to think about how AI tools are amplifying and adding to human potential. Contrast what we are able to do now with, say, Project Aristotle — Google's research on what made teams effective. At that time, we looked at about 400 different variables of teams across 180 teams over 18 months, based on surveys and static information from HRIS systems. That is very different from what we're able to do today with tools like Nadia. Carrying the gym analogy forward: it's like asking about your workout through a survey versus having an Oura ring. We're now able to get into the depth of not just someone's thinking, but the quality of their thinking process. That is what we can do today with AI coaching tools.

Das Rush: One of the things that came out of what you created is this idea of a Power User Index. This isn't necessarily what you set out to create, but as you dug into this problem, you realized it was really important to find a new kind of measurement — a way to understand what AI adoption looked like. Talk me through what exactly the Power User Index is, Prasad, and why you created it.

Prasad Setty: What we set out to do was: when we have an amazing tool like Nadia and want it used well, what information accounts for good usage? We wanted to look at several different variables that go into what good usage looks like. Certainly the volume and frequency of use, but also the breadth of conversations — are people thinking about it for different things? Goal setting, feedback, onboarding. And then the third is conversation quality. AI is so capable of giving answers that it's easy to get a quick response. Some questions are 'should I do X or Y?' But others go much deeper — what are the risks? What are the second-order effects? We wanted that conversation quality factored in as well. Thanks to Valence's data sciences team, all of this was quantified into an algorithm that allowed us to create a much more powerful and compelling usage index.

What Is a Power User Index for AI Adoption?

A Power User Index measures the depth and quality of AI tool adoption — not just whether someone logs in. Developed by Prasad Setty in collaboration with Valence's data science team, the index combines three variables: frequency of use, breadth of use cases, and conversation quality. Power users regularly engage with AI coaching for high-stakes decisions and explore multiple use cases, while casual users open the tool occasionally and ask surface-level questions.

Prasad Setty: A casual user is someone who sees an email that it's performance management season, wants to save one hour, uses Nadia for it, gets the hour back — and then forgets about it because that's the last time they saw the email. A power user goes there regularly, thinks about all those kinds of high-stakes conversations, and says: 'For these decisions I'm about to make, I need to have the best perspective. I need to think about it from multiple perspectives.' That is the texture of what distinguishes power users from casual users.

AI Coaching Power Users and Performance Rating Improvement

Das Rush: There are two different things that happened as you worked through the data. One was separating out what power usage looks like versus casual usage. But you also banded performance — low, medium, high performers. You looked at: if someone was a low performer, were AI power users more likely to move up a band than their counterparts? Jennifer, what did you find comparing casual and power usage?

Jennifer Carpenter: One thing I would say to all the talent leaders in the audience: the only people who are going to figure this out are us. We don't have answers, but we have the questions we need to be curious about. I truly believe we are guardians of growth — growth of our companies, growth of the people in our organizations. We need to figure this out.

From a performance perspective, those who were power users — using Nadia a lot and for a variety of use cases — were 

28% more likely to move up into a higher rating category.

When we looked at who was using Nadia in the top 30% performance tier: casual users were represented at 37%. With power users — or even just high-frequency users — it jumped to 47% in the top rating category. Now, you can say: well, your top people are the more curious ones, the ones who lean in. That's correlation, not causation. Fair. But that's why we looked at the moving up piece next.

Jennifer Carpenter: The second finding is really important too. When managers are power users, the people on their teams are two times more likely to be users of Nadia. This is not an AI problem — it's a leadership problem. What managers do and say is what their teams are going to do and say.

How AI Coaching Power Usage Correlates with Performance Improvement

Employees who became power users of AI coaching were 28% more likely to advance to a higher performance rating, according to research by Jennifer Carpenter of ADI and Prasad Setty of Stanford, analyzing 45,000 sessions on Valence's Nadia platform. The study used a natural experiment: it examined only employees who began using Nadia between two performance rating periods, allowing researchers to observe before-and-after performance shifts. The association is strong; full causal proof requires further study.

Prasad Setty: What Jennifer and ADI had was a natural experiment. We had prior-year performance ratings — that is why we can see the shifts. And we only looked at people who started using Nadia between those two performance periods. It wasn't a completely randomized control experiment. Those will follow in the future. We are strongly suggesting a great association. We are not yet fully explaining causation, but we are very encouraged by these trends. Of course, there were fewer power users in the low and mid-performance ratings initially. But if they were power users, they had a higher chance of moving up. That is the power of power users.

How Manager Adoption Drives Team-Wide AI Coaching Engagement

Das Rush: You touched on the role managers play in all of this. I want to take a moment to allow you to double-click on what the data has suggested about managers adopting AI coaching — and what it looks like when they're power users.

Jennifer Carpenter: Across ADI, about 65% of our employees have tried Nadia, 77% of our managers have, and 61% of individual contributors. So managers are more likely to use Nadia than individual contributors, but still healthy and strong engagement across the board.

I'll say it once, I'll say it twice, I'll say it three times: leaders influence adoption. If your manager uses Nadia, you are engaging at 67%. Those whose managers aren't using it are only at 33%. I was shocked to see it that stark. That was something we discovered as we unpacked this — and it was a very surprising insight.

One more finding on managers: those with higher spans of control — ten or more direct reports — are engaging more often with Nadia than those with smaller teams. And as Ethan mentioned this morning, it's likely that as our work changes, we're going to have larger spans of control. AI coaching is another way in which tools like Nadia can support leaders as they have broader impact across their organizations.

Why Manager Behavior Is the Strongest Predictor of Team AI Adoption

At ADI, teams whose managers were power users of Nadia AI coaching had a 67% engagement rate, compared to just 33% for teams whose managers did not use it. When managers are AI coaching power users, their direct reports are two times more likely to become active users themselves. Jennifer Carpenter, ADI's Global Head of Talent, frames this plainly: AI adoption is not an AI problem — it's a leadership problem.

Prasad Setty: One of the implications I've been thinking about: when we measure span of control currently, it is usually the number of direct reports a manager has. I think there's a corollary metric that we'll also start measuring — for lack of a better word, I think of it as span of work. As independent AI agents do work and exercise authority, leaders may not expand their span of direct reports, but they will expand the span of work under their purview — the span of things they are accountable for. That has to go into organizational design as we think through the future.

AI Coaching Across Geographies, Generations, and Gender

Jennifer Carpenter: We're a global company — headquartered in the United States but across Europe and APAC. You might guess the U.S. has the most engaged users. You'd be wrong. APAC employees were 15% more likely to engage with Nadia than U.S. employees, and EMEA was 6% more likely to engage. I thought the U.S. would be our most engaged users, and they were not.

You'll be as excited as I am to learn that age really doesn't matter. Employees under the age of 30 are only 10% more likely to use Nadia than those over 50, and I saw no statistical significance in the other age groups. That's an optimistic insight.

Let me stop on gender, because as talent leaders, we need to be asking: am I making inequities better or worse? Back when we rolled out Nadia, I made sure the first groups of thousands who had Nadia had 50/50 gender parity — which I might not have done if I had simply said, 'Let's give it to all the managers.' Because we know managers are not 50/50 male-female.

Studies show women are less likely to use generative AI when controlling for all factors. What we found after 14 months: among casual users, our population is roughly 30% female, 70% male — slightly higher than the 26% female representation in our broader ADI workforce. But when we look at those power users — the top 5% — it jumps to almost gender parity: 47% female, 53% male. That means women are 

68% more likely to be in the high-frequency power user group relative to their share of the workforce.

I'm still trying to wrap my head around what that means. I'm proud of it because it means women are finding value and leaning in during these early innings. That's what we need everyone at our companies to do: get in, find the value, find the use case.

How AI Coaching Can Support Gender Equity in Leadership Development

Despite research showing women are generally less likely to use generative AI tools, ADI's 14-month study found women were 68% more likely than their workforce representation to appear in the power user tier of Nadia AI coaching. Jennifer Carpenter attributes this in part to ADI's intentional rollout strategy: ensuring the first cohorts included 50/50 gender parity, rather than simply distributing access to managers — a group that skews male in most organizations.

The Workforce Bifurcation Risk: AI Haves and Have-Nots

Das Rush: One of the most provocative ideas here is workforce bifurcation. If you have high performers who are more likely to adopt an AI coach, and as they adopt it their performance accelerates, then AI haves and have-nots isn't just an organizational story — it's going to happen within organizations and their workforces. How should leaders, given this data, be thinking about creating power users within their organizations?

Jennifer Carpenter: Three things I'm following as a mantra. First: who are you enabling? Lead with equity. Make sure you're not unintentionally enabling one group that will further bifurcate your organization — which I almost did when I said, 'Let's just give it to managers.' Give it to everybody.

Second: invites, not edicts. Invite people to lean in, invite people to try something. Don't force it. We had a small rollout initially with GitHub Copilot — I thought everyone would throw a parade, but they weren't thrilled about being mandated. People don't learn when they're on their back foot. They learn when they can lean in and feel engaged and invited.

Third: find the pain to get the gain. Nobody likes writing self-reviews or setting goals. If a tool like Nadia can lift mental load off someone's shoulders, it helps people engage — and then they can see the art of the possible and try the next thing, and the next.

Prasad Setty: What I love is the comprehensiveness of how you're engineering the system to increase participation — because with something like the Power User Index, there is no fixed pie. Everyone can be a power user. Let me work through some math on why bifurcation risk is real.

Say you have Bob and Tina. Bob produces 100 units of productivity, Tina produces 120. Now, AI amplifies everyone's output — say by 20%. Bob is now at 120, Tina at 144. The absolute gap has already increased from 20 units to 24. Now assume power users get a bigger amplification: Bob at 20%, Tina at 40%. Bob is at 120, Tina is at 168. That is how bifurcation stretches organizations. All the social engineering Jennifer described is key to making sure we don't end up with a group of haves and have-nots.

What Workforce Bifurcation Means in the Age of AI Coaching

Workforce bifurcation refers to the growing performance gap between employees who deeply engage with AI tools and those who do not. Stanford researcher Prasad Setty illustrates the compounding math: if a power user gets a 40% AI amplification while a casual user gets 20%, an initial 20-unit output gap grows to a 48-unit gap — without any change in the employees themselves. HR leaders who do not actively engineer equitable AI coaching adoption risk accelerating this divide inside their own organizations.